The Adam Sand Report - Thursday, October 1, 2026

What's going on everybody? Today we are excited to be talking about a little bit different topic than we usually do. I want to talk to you guys about robots and robot news that you can use and how you can understand what the coming impact is going to be. Now I will probably cater this mostly to roofing because for those who don't know, I am working on a robotics process for getting robots into the roofing space. And this is a, a long-tail problem to solve. You don't solve it in weeks, months, a year. It is a multiple-year plan. And essentially the thesis is, is that robots will work on roofs, and this will happen in 3 to 5 years. And in 3 to 5 years, to do that, you need at least 400,000 training hours to train the robot. And in order to get the training, you have to get roofers to wear training gear. And in order for roofers to wear training gear, you have to pay them. And in order to pay them, you have to own roofing companies. And therefore, the goal is to own roofing companies. Now, to get investors to be interested in owning roofing companies, well, they have to have an interest in short, medium, and long term. So the The short term is the risk is reduced by owning the roofing company. The medium term is you reduce the expenses of the roofing company through deploying something that's tangible to them today, which is the Agentic Operating System. So this is the AI roll-up concept that's been popularized by Thrive and General Catalyst. This is talking about, you know, buying companies and like a venture capital firm would buy them, not like a private equity firm would buy them. with the intention to take them public, reduce the expenses, improve the EBITDA through the Agentic operations, and then long-term secure the ability to make the training dataset. Now, to understand the training dataset for roofing, what is behind this 400,000 hours? Well, that is the moat that you can put around the business that secures it for the long term. Now, if you're going to— if you're going to try and get something like that done, Well, if you're— you have to give a stake to the roofers because the— if you look at this AI and the software developers and everything like that, they're super depressed, super upset with the fact that, you know, AI has come into their world, um, because it's like there is this, you know, growing feeling that they're being replaced, right? That there is, uh, that there is like no opportunity left for them. If you read this post right here, this right here, and go like this, put this on stage. Okay, if you look at this right here, well, this developer says, hey, I'm done with this, it's over. The state of engineering right now is horrible. It has been half a month since I started a new role at a big company. Nobody knows anything here. The specs, the code to test, the PRDs, the tickets, the resolution of those tickets, reports, etc. Everything is made by clogged code. Nobody on my team likes this. They're being forced to ship as much as they can. I've heard multiple times from higher management that pushing code is not a bottleneck. So why are we so slow? People are working 12 to 13 hours a day just to press enter. Nobody is reading anything. Humans in corporate are doing nothing on their own. Everyone literally Where was I? Everyone, literally everyone from an L1 to an L7 engineer here is doing the same thing. Talk to Claude. There's no sense of victory. Nobody is resolving bugs. In reality, nobody is thinking anymore. Everything is done by LLMs. It is so soul-sucking. I would not mind it, to be honest, if we were at least given the time to check out the code and see what is going where. But no, the goal is just to ship no matter what happens. Now I will say that this is not uncommon. That the reason that I was showing my screen like that was because I sent that message to a friend of mine who works for, um, like, you know, multi-billion dollar 3-letter acronym that I guarantee you everybody in this room, no matter who watches this video live or afterward, has heard. He's a principal there, and, uh, this company has, you know, 40,000 engineers. And I just said, how much of this is true, right? Um, and he said that, you know, it's— this is partially true with that company, um, but it depends on their passion, right? Because the passionate ones are, are finding their way to be useful in this new world. But nonetheless, it is a concern. And the problem is that they feel like they're being pushed to productivity, but there is no way to measure productivity anymore as a software engineer. And I promise this is going to come back to roofing and robots in a minute. But you have to understand this narrow view first. The first place where AI kind of took jobs, right, was ironically like the first, I guess, volley of attacks. The first, you know, fires across the bow of the ship was graphic design. The very first model that anybody in the general public got access to was actually DALL-E and DALL-E 2, which is graphic design, but it sucked. We all know it sucked. And then you had Midjourney, and then you got the very first video generators, and you had the Will Smith eating spaghetti video. And then everybody kind of said, see, AI sucks. And then we got ChatGPT, and then we started using it in Copilot and this thing called Replit. You have these tools for auto— autopilot, or sorry, autocorrect in code. So you could type some code, it would kind of, in blurry text— not blurry, but like faded text— show you what the end of that particular function would be. You can hit Tab, Enter, and it's like autocorrect for code. Right? And it felt like spell check, so it felt non-threatening. But what happened was AI downloaded all of code, and because code is deterministic and images are probabilistic, right? It's probably going to look like that, but code is deterministic. It either works or it doesn't. You run it and it works or it doesn't work. And so there was these sites like Substack and whatnot where you could go to get coding advice. It was mostly just jerks. And so when something came out that you could be like, hey, what do I do here? And it wasn't mean to you, and you'd have to deal with all the Like think of YouTube comments on Substack. That's what it was like. If you wanted help coding, it was just getting ridiculed. You should know this. I can't believe you're this. You say this in your profile, and I can't believe you're asking this question. Anybody should know how to center a div. Well, in coding, it was the first thing that was kind of taken away, right? And everybody just let it happen, right? AI just stole the entire world's knowledge, but only a few pieces of that knowledge is deterministic. Things like math and coding. And the first economically valuable one of the two was coding. So if you want to see, like, so goes coding, so goes the world. And there's no way to measure productivity in coding because if it's lines of code, right, if it's just lines of code, well then, I mean, Claude will write 6 lines of comments inside of code because your documentation is no longer documentation. Your documentation is the code and the comments inside the code. But then your code could have— A +10,000 diff and 4,000 of it is just comments that Claude is leaving behind to leave, you know, trail of breadcrumbs. So it's not that. Well, then maybe it was like bugs caught, right? But then what the incentive is to write code with bugs so that they get caught? Or is it bugs escaped? Or maybe it's contributions on GitHub, but you can make— you can get a contribution just for clicking through and hitting review and not actually reviewing it. And you can just, you can just, uh, you can not squash and merge, you can just merge, and then you'll get more contributions because you'll get all of your commits as, um, as, as merged. And then you can fake the contribution. So then maybe you say it's PRs, but now with, you know, these coding agents, you can now put so much into a PR that, you know, now you can have a 20,000, um, you know, like, like, you know, plus, minus, green, red, you get like a plus 20,000 you know, 17-commit, 43-file PR, and you can also have like a 2-file PR, right, where like 4 lines have changed. So it's not PRs, it's not lines of code, it's not contributions, it's not bugs caught, it's not bugs escaped. And then you just get to this like amorphous term of like shipping value or like a global thing of like features closed that customers asked for. But then now there's this influence on the code to say, well, you know, nobody's reviewing the PRs because it takes you— the human is now the bottleneck, right? It's like if you look at that, um, oh, what did they call that thing? I always forget the name of it, but the, um, Navier-Stokes, the math problem that OpenAI solved. They spent $8 million to win a million-dollar prize. It was a 100-year-old math problem. They— it's not technically solved yet because it takes 2 years for it to be reviewed. But that Navier-Stokes problem, there was 4 criteria to solve it, and you had to solve 2 of the 4 to consider it like worthy of winning the Millennium Prize. It did solve 2, but to qualify, the response is 13 million lines. So like, who's going to review that? And to accomplish it, they put a swarm of 10,000 agents on it for essentially a weekend.

Right.

And it equated to 100 years of human work. So how are you going to— when you put that kind of power behind a coding agent, who's going to review the PRs at that rate? You're going to get AI to review the PRs and then you're going to just merge it. And so this is creating this environment of a lack of fulfillment in the AI coding space because you're just hitting enter, you're just approving, you're just saying yes, no, yes, no. You're You're not figuring out and tinkering with the code to try and solve the problem. And so this is creating this like depressing state. Juniors can't get jobs because frankly, Claude is the equivalent of a junior engineer. So the senior engineer could just write the spec or coordinate the creation of the spec through something like a Wayfinder to a grilling session to prototype to spec to tickets, and then run it all linear and put it in the software factory. You don't need the junior. Anymore. And so juniors can't get jobs. The ones that do have jobs feel like they're just hitting enter all day. And the managers who are saying, well, what's in this code? They're like, I don't know. Like, well, what did you write?

I'm not really—

uh, hold on, let me ask my agent. So it's like, well, I could ask the agent that. So then it just feels very circular. So what the hell does this have to do with robots and roofing? Well, what hap— right now the cost is fairly high, but the cost is getting more efficient. Down to the point where the, the model that kind of broke the internet, or not the model, but the product that broke the internet about a year ago is OpenCLAW, right? The concept of the agent harness, right? The concept of an agent that had kind of a set of instructions that it, that it wasn't just something that has a trigger and then you had an LLM and then maybe some tool access or an MCP, but actually had a personality and identity and memory and the ability to like do long-running tasks and to remember what it talked about because it had its own localized memory. That was New. Well, the model that was crazy back then was Opus 4.5. Well, today Opus 4.5 is completely different. Watch this thing from Tom Bilyeu where they actually talk about this.

Like, these AIs right now actually have pretty much reached the efficiency of a human brain.

Says who?

From an energy perspective, they're way worse than a human brain.

Way worse. Nope.

Just say more.

What have you seen? Because I haven't heard anybody make that claim.

So this is actually a really, really smart mathematician, has his own AI startups. He's highly respected in the industry, been very successful. His opinions are considered to be expert.

So if you look at the latest Qwen models, right?

Yeah.

So you know Qwen 27B?

I know of, couldn't tell you anything about them.

So the model that kicked off the OpenCLAW thing at the end of last year was Sonnet 4.5, Opus 4.5. There was state-of-the-art at the start of this year. There is a model from a Chinese Alibaba called Quen27B. It's a 27 billion parameter model. What does that mean? It's like 60 gigabytes. And so you need 60 gigabytes of RAM to run it as default. They got it down to like—

So think of that, 60 gigabytes. That's 100— like you can put that into a 96 gigabyte Mac Mini. But let's keep going.

25. A team called PrismML got it down to 6.

Same model.

Same model.

Just some sort of more efficiency.

Yeah, yeah, they got—

they moved 6 gigabytes around. So that's a MacBook Air or a Mac Neo, or I think they call it Neo, like the new— basically it's an iPad with a keyboard.

Kept it from being 16-bit numbers to 1.5-bit numbers.

Pretend I know how that plays out. Sure.

So like, I don't know what he means there. It's called quantization. Quantization. So they had 16-bit quantization to 1.5-bit quantization. That's what he's saying.

So if you want to Google the technical word for that, quantization is rather than having 16 bits of different numbers, they managed to magically squish it and retain 95% of the performance. And this model matches the Frontier Opus model with that degradation that you were using in January.

So, Opus 4.5 from 1, 1.5-bit quantization in 6 gigabytes.

And it runs on 6 gigabytes.

Damn.

Which is a Raspberry Pi, which uses as much energy as your brain.

Wow.

They just announced a 1-bit version at the Snapdragon conference today. that runs on your glasses. And again, think about the Opus you were using, the Anthropic model in January. That's a pretty smart model. It wasn't crazy smart, right? But was it reasonable?

Yeah. Yeah.

Again, was it?

So if we're getting to the point now where we have on our glasses Opus 4.5, if we have that level of intelligence when we're looking out at the world, That is going to drastically shape the way that human productivity is viewed because humans will become the bottleneck. Humans will become the slow thing. We're— this whole video at the beginning, the, the mathematician postulates that we as human beings in knowledge work will be the slow, dumb thing in the middle. So right now we have human in the loop, human on the loop for the purpose of like fixing the AI. But today, these Frontier models, like most normal people walking around, if they hop into Opus 5.5 today, they don't even understand how to use AI to actually take advantage of it. They don't need Frontier Intelligence. They would be perfectly fine with the Opus 4.5 that is in their smart glasses. It's more about the, it's more about the, the, the place that they can use it, the, the function, the modality, the, the The utility of the AI, where it is in their life, not that it is frontier, which is the case that he's making here. But then we get into talking about the robots, right? We get into talking about what does it mean for robots to be walking around. I got some interesting news for you here. So this is something that is pretty exciting, is this new farmhand robot. So you are seeing robots come into what you would consider to be the trades. There are so many different robot robotics companies coming out right now. It started— you saw a lot of these Boston Dynamics robots, but you're seeing what does this look like in a farm application, right? The ability to monitor your plants. And just like you're thinking, that Quen 27B model that runs on 1.5-bit quantization that can do OPUS 4.5 level work, and it can work right from A simple camera mounted on a robot and it can go autonomously monitor your plants. But then also you're seeing humanoids that could keep track of livestock, that can be monitoring things so that, you know, you're not going to hear what the cow's getting out, right? Can make sure the gates are closed. It can, it can make sure that there isn't any, like, you know, I don't know, I don't know nothing about ranching, but like, I mean, I've seen Yellowstone enough to know that, like, what happens if the coyotes and the wolves get in there, right? You have something there that could monitor it, that could shut the doors, that could monitor for the weather and close the doors, open the doors, monitor the temperature. Or understand if there's like a birth happening and I guess somebody could run in there, or some kind of appearance of sickness or something like that. There's some— I can monitor the environment. So this is how robots find their way into the real world. And like I am saying that you got to put them on roofs, but this is mostly just a showpiece. For it to actually be useful, there's a couple things that have to happen, right? So you have to think about the onboard AI capabilities. You got to think about battery life, and you got to think about the Electromagnetic tensile strength of the hands. Well, what just came out? Boston Dynamics has a brand new hand. So this is the big news of today: is this new hand that has this incredibly dexterous, mobile, strong, flexible. It has more degrees range of motion. It can do incredible things. Stuff that so what we're seeing today is the new generation of hands for Atlas. They're a perfect companion to where Atlas is today. It's been designed with simplicity in mind. By the way, why is everything Atlas? I named my kid Atlas 3 years ago and everything ever since then has come out with the name Atlas.

Trajectory to mass manufacturing.

For the kinematics of the hand, we have increased the number of degrees of freedom. GR2 had 7 degrees of freedom. GR3 now has 13. The size and strength of the new hand is a little bit larger than an average human hand, but it's also on average a bit smaller.

Any forces that apply to the environment gets applied very transparently.

Proprioception.

So what happens when one robot manufacturer solves battery, another manufacturer solves onboard AI, another manufacturer solves back strength, another manufacturer solves hand strength, and these things become modular? These things become inter— interchangeable.

Backdrivability is not only how much force is applied to Emily's golf balls. Backdrivability is not only—

this one is kinematics— Atlas is going at it from one direction, but approaching it from both directions.

We can see that and spend time here for that.

Worth the assumption.

What do we have to do today? The multi-fingered hand, it's like a whole mini robot when you think about the number of joints and the number of motors in there.

The hardest thing that we were working on This gripper, I was not expecting to do so much science in this.

So obviously every single part of this requires innovation to come to it, and these robot factories are moving fast. Now for a little bit of fun, if you like Terminator, because a lot of people whenever I post robot content they always talk about Terminator, well Figure, who has launched Figure Index and they have their new Helix 2.5 that they sent around working people's homes, I've, I've seen incredible footage of this robot basically working, working around home. So this is a non-teleoperated— so this is onboard AI looking around at its environment and figuring out its environment and knowing what tasks it has to do: picking up stuff, putting it away, putting stuff in the laundry, picking up toys, putting in the bin, making beds. And these, and these robots were put into 30 rental homes. They had no additional training, no context of the place they were being put in, and it was just people going around messing up the house, and then they would have to go back and, and clean the house. And this robot is doing this, making these beds, and this got a ton of views. And now they— so they just decommissioned their previous round of robots. Bigger, bigger dudes. This is your Terminator scene. Look at that. What movie is that from? Oh, the man himself. Like, these robot companies are having a lot of fun. Look at that, the iconic thumbs up. And again, these were not— I'll let you get it. Hasta la vista, Figure 2. These were not teleoperated. These were trollers. They're trained to perform these stunts where they actually run and jump in, backflips and front flips and stuff. Well, this one's not backflips. Pretty crazy, right? So you're seeing the decommissioning of the old ones, and if you're wondering about the volume of production, well, Tesla just announced that it's actually moving its production. Um, it is actually taking the chips that they were putting in Tesla cars And they are now putting in this RAM into the Tesla Optimus because it was the only way to keep up with the level of production. And so I actually asked Grok here to see what it would say. It says no official Optimus production targets. Reports indicate hundreds per week now at Fremont, targeting over 1,000 weekly by end of 2025— or sorry, end of 2026. So the production on this is going to go from nothing to all at once. And this guy actually talks about it at around minute They're not quite there yet.

The next generation—

They freak me out right now.

Yeah.

So last year, Unitree, which is the very famous one, like, it's called the empty face.

Yeah.

You see them everywhere.

Yeah.

Yeah.

They sold 11,000 units.

In a year? In a month?

Overall.

Lifetime?

Lifetime of those humanoids. Pittance. Yeah.

But it's happening.

Yeah. Yubitech sold 16,000 on one day of their human-realistic ones, and the max price was $165,000.

Whoa.

So they're starting to roll those out now.

Okay, I'm going to ask Avi's question. Are they sex bots or?

Apparently, you can customize them any way you want.

Hey, okay.

Like, people are lonely. What can you say? But I mean, you're going to see tens of thousands of these, millions of these coming. They're going to be everywhere that you see. And again, like, if I said to you, Tom, $30,000 gets you a humanoid that can cook like a cordon bleu or Michelin star chef, Yeah, no, that's wild. And like—

100%.

100%.

Dude, legitimately, I'm like placing my order as soon as the episode is over if that's actually true. That's crazy.

But again, you look at the manual dexterity of these things now, like everything's broken through on those levels. But then what does it mean when you start ramping them up? You can only make 70 million cars and 70 million motorcycles a year. But I tell you now, currently, every factory in China is moving fully dark, no humans. Trying to build these robots as fast as they can.

And when you say fully dark, no humans, because the robots themselves are assembling the robots, so there's no need to pay an electricity bill for the lights.

Yeah, so Xiaomi now doing electric cars and robots. Everyone's doing robots.

So this is coming. And right now it's interesting, Anthropic did a study about the impacts of the labor market. There was this one on March 5th, and then this one that they did more recently. It's a little more visual. But it basically talks about what is the impact of the labor market on, um, our— as a result of AI. And so their research study says that the economy is made out of tasks, and this model represents all the jobs people do in the economy as a bundle of tasks. And this is how you can look at your business, whatever business you're in. The open and closed-loop workflows, right? And then attach the tacit knowledge and business primitives Right? So that this is how people work. And so AI can help people do a given task better or faster. It can automate the task. It might not affect the task at all. It can lead to— and it can also lead to new tasks. And so think of a day in the life of a nurse, right? They have all of these jobs. These are bundles of tasks that they do. So triage the incoming patients. This is triage the incoming service tickets at a software company, or triaging the incoming tickets at an IT services company. Right. Or like any kind of thing like that. There's handwrite the paper chart, chart, chart their vitals, chat with the patient, bathe the patient, plan the shift's care schedule, do the rounds, walk the paper notes to records, order supplies for the ward, draw blood, call in the supply order by phone. Right. Coordinate with the doctors, count the ward stock by hand, draft the discharge instructions, schedule the follow-up visits, and train a new nurse. Well, all of these jobs tend to change over time, right? You'll have more new ones, you'll have old ones, you'll have ones that aren't new and aren't happening anymore. Because like 30 years ago, no one monitored patients remotely, right? They didn't have that. And only humans can do some tasks, but AI can do and augment some of those tasks. And that's what's represented by the purple here, right? Like planning the care, the shift's care schedule, coordinating the doctors. Drafting the discharge instructions, running the ward's dashboards, right? All of those things can happen by AI, and some can be fully automated, like ordering supplies, charting vital vitals, and schedule the follow-up visits. And so these, these will now create new tasks like check how well the AI triaged and review the care plan the AI proposed. So there'll be new jobs that come, but the nurse's job will change. Some of these boxes will change size, some will get bigger. Now, this is the same thing for your business, and this is Why it's so important right now to be absolutely obsessed with this, because if you do not, you will get beat by companies who don't try and just buy an AI widget or an AI task doer or an AI service, because the rate and the rate of this has gone so fast. We started with just AI autocorrect, right? And then we went into AI autocomplete. And then we had, you know, the concept of like models that you could go copy paste things in and out of. And then you got access to the API and everything was basically a ChatGPT wrapper. That was the thing that a lot of venture capitalists like poked at AI startups for. It was like, this is just a GPT wrapper. You're just basically, instead of copy pasting something into ChatGPT, you're automating that. And then you saw every business was starting to become an N8N, or not every business, but every, every agency. There was all these N8N agencies that started building agentic workflows or AI workflows where you had a trigger, delivered a package of data, you had an LLM, it could have a structured output, it put that structured output out, and then you had another workflow step. So it was just Zapier with AI in the middle. And then you had agents that could actually house and go back and forth. So then you started seeing everybody selling AI chatbots for their website or AI assistants. Then you got, then you started to get like these actual agents, these things that had an LLM, a memory, and tools, and they could work autonomously. They could start up and stop on their own. Then you got into the agent harnesses, and then you had the OpenCLAW moment where everybody came out and you had these OpenCLAWs. Then you get the OpenCLAWs for dummies, because after OpenCLAW Hermes, then you get the GroqBox, the MetaMuse, and now the OpenAI Dots. And so people started building AI startups that were basically taking a highly niche, highly problematic, usually complex task. There's 5 qualifying criteria. It had to be— it had to happen often but intermittently. You didn't know when it was going to happen. It had to be relatively low cognitive load. And that line, that line of cognitive load has gone up over time. But it had to be a high cost of inaction with disparate systems that added complexity. So 2 different systems. So if you were trying to be an AI startup 12 months ago, you just had to find something in a business that happened often, but not on a routine schedule. It was, you know, it was something that could be deferred to AI and call that cognitive load. And then it could be, but the cost of not doing it was high. So there was an expensive ticket or price to pay if it wasn't done or wasn't done on time, and it required the use of disparate systems. That was an AI startup a year ago, and you got all these niche things that would do document review or policy compliance or something like that. And so then These companies would buy 30 of these endpoint solutions and companies would build 30 of these endpoint solutions and then they eventually become this agentic operating system. But then it becomes so easy to build something like an agentic operating system and you're always under threat by Claude and OpenAI's Codex to basically be removed from being needed because now you can connect knowledge bases and plugins and skills and all these kinds of things. And then now you move into this AI rollup, which is essentially the people who build the AI. Operating system have no reason not to just become the business. And this is where Thrive and General Catalyst came in because they said, well, selling enterprise software is a real pain in the ass. And so you see a company, an AI startup that would build this enterprise software for, for private hospitals that could make private hospitals more profitable, and trying to sell it to the hospitals was hard. So then they raise a $9 billion fund and say, go buy the hospitals. And the same thing happened in IT managed service partner space and in the accounting space for accounting firms where they just say, you know what, just buy the accounting firms and deploy this technology into it, bring down the costs, improve the EBITDA. And this is a different— this is a venture capitalist approach to a roll-up. And so what happens as these jobs change, the problem is they happen millions of times every day across the entire country and they can all happen at once. And these tasks start to improve the economy And so you'll see that all of a sudden the GDP will start to go up. And there's different examples of what this might look like. But there's always a but. Even in the modest example, what happens is that knowledge work starts to become an advantage for those who know how to use AI, but becomes a disadvantage for those who don't. And so then what happens is you see that there's these Skills lead. There's these levels of AI adoption that are happening. But what is— what you're going to see here is then knowledge workers are expected to be displaced, right, into all other workers. And what is all other workers? It's basically saying accountants are going to have to go become plumbers. And you're seeing this. You're seeing university graduates drop out of university because getting a job as a junior at Microsoft or Facebook is almost impossible today. But the, um, but the, like, you know, being a plumber or an electrician right now is blowing up because the construction of data centers is making that such a successful job position. But not every accountant wants to necessarily go become a plumber or a roofer. And so you're— they're essentially saying, don't worry, it's not that big of a deal because yeah, we're going to have all this like displacement of knowledge workers, but they're going to cross over. Well, what did we just show you? They're not going to cross over because by the time they try to cross over, even the surgeons are under threat because this is a robotic surgeon's hand. I haven't showed you this yet, but this here is showing how surgery is going to be affected. This is a new startup funded by a16z, which has a whole— it's a fund of funds of billions of dollars of funds that are investing in robotics.

We are building a self-improving system. A system that is capable of learning and ultimately performing any procedure.

And this is not new. I mean, do you want LASIK, some guy with his hands doing laser on your eyes, or would you rather have the LASIK machine? This is just the LASIK machine brought to everything.

We are training large behavioral models on massive amounts of surgical data. There are procedures that only a few hands on the planet can perform, and so people just don't get access to the surgical care that we need. Only now do we have the tools to begin attacking this problem. Fundamentally, like, these systems need to be built from the ground up for autonomy, and that's what we're doing.

So, how do you think this technology can help So this, so this whole concept of like, oh well, they're just going to go become welders and plumbers, that ain't going to happen. Because what they don't take into consideration in this report is the impact of robotics on the workers that are going to be displaced. And so this is the time now more than ever to become the— you have to own or be owned when it comes to the labor force. And this is why we are building foundation project. This is my— well, this is my project essentially saying we're going to do this in roofing, which is that roofing companies are going to be integral part of this. And it's not going to happen all at once. We have to partner with them. We have to make sure that this business focuses on the people, the data, the systems, the playbooks, the training. We implement the sales and production frameworks that we know work. We create the marketing accountability because you can't without leads. We automate, bring in the financial operations, use the data intelligence to create an agentic workforce, add procurement leverage, employee stock ownership for the long term, eventually IPO, which is how we grow the business. And then we get the robotics integration, and this is where roofers can exit their business, but also employees can have a stake in the long-term success of the stock because they will have shares in the training data that eventually replaces their jobs. This is the whole purpose of what we're doing. So to train the robots, we need the roofers, and to get the roofers to do it, we have to own the roofers. And so we're partnering with roofers to be a part of this robotic agentic future. This is what I'm very excited about, and this is why I want to educate people about this and get the awareness out there. Because for me, I don't see a future where you don't take— where you don't become a part of this. And it just goes away. And I have a lot of feeling for the humans on this. If you happen to have watched to the end of the video, this is where you can know about this little secret, this little secret A down here. And if you click it 3 times, you'll get to see the secret backend, which is my AI manifesto, which is the months and months and months of me thinking about this, which culminated into this big long writing project, right? This conversation that I actually had with AI to curate all the thoughts that I had turned into turned into some points that I eventually turned into some Twitter-level sound bites, and then I turned them into videos. You could find them on our Facebook page. But thank you very much for watching. It was fun to have you guys here. I am— I do these live streams just to find my people, to find the people that I would connect with, people have the similar ideas. That's why they do them late at night, is I want to create the opportunity to just document the journey. I'm not worried about getting a ton of views, but I want to basically put this journey out there, and that way, you know, 2 years from now when all this starts to look like a foregone conclusion, we can say we were here and this is what we were doing. All right, have a great night.